Machine Learning Classification Pipeline

This project is a two-part machine learning pipeline demonstrating the use of both tabular and image data for classification tasks.

Part 1: E-Commerce Revenue Prediction

• Dataset: Preprocessed shopping.csv file representing user behavior on an e-commerce site.
• Goals:
  • Predict whether a user will generate revenue.
  • Compare performance of custom linear regression (with and without regularization) and a deep neural network.
• Techniques:
  • Data normalization: Min-Max, Z-score, and Mean Normalization.
  • Manual implementation of gradient descent for logistic regression.
  • TensorFlow/Keras neural network with 4 dense layers (last using sigmoid for binary classification).
  • Evaluation metrics: Accuracy, Precision, Recall, F1 Score.
  • Tested on a separate dataset.

Part 2: Traffic Sign Classification (Image Data)

• Dataset: German Traffic Sign Recognition Benchmark (GTSRB).
• Goals:
  • Classify traffic signs into 43 categories based on image input.
• Techniques:
  • Manual image loading and resizing using OpenCV.
  • Image flattening and normalization.
  • Deep Neural Network using TensorFlow/Keras:
    • Input layer: Flatten
    • Two dense layers with L2 regularization and ReLU activation
    • Output layer with softmax activation
  • Model trained on 60% of data and evaluated on the remaining 40%.